Welcome to Idle Entity.net

This is the base of operations for Graham Lord.
From here I broadcast a mediocre blog of events that interest me, and the slow development of my on-going digital hub.

Latest Blog

2010-07-09 11:49:03 / On Robots and AI

So I was watching I, Robot with a mate the other day, and it got us onto an excellent conversation about AI and the likes.
I'm of the opinion that we will develop AI, but the first AIs will not be the self-aware computers / robots hollywood envisions.
First we have to consider that while we'll invent anything we can think of, nothing is produced on a small or mass scale unless there is a practical use for it. Take Quantum Computing as an example, this has been moving at a relatively slow rate in recent times, because we've proven that we can build quantum computers, but at present, there are little or no practical uses for them that a binary computer is not better suited for, which is why we're not seeing any quantum functionality hitting production computers.

With that in mind, we ask our selves; What use have we for a self-aware AI, or even an AI that can think for itself in any way?
The answer is none. Thinking computers are only going to bring us trouble, because of all the moral implications (robot rights, etc) and the safety measures we'd inevitably have to take to keep it under control.

So what is a useful AI? In my opinion, it's not a computer that thinks for itself- it's a computer that understands humans. All computers we currently have speak the language of computers, and that's it. If you want to interact, you've got to learn the language. That might be the programming language the computer runs on, or it might simply be moving and clicking the mouse. Even a command line isn't in english- you can't type "Open my shopping list" into BASH and expect to see you vegetables listed in the order in which they're arranged in the supermarket.
So what we need is a computer intelligent to understand humans and do s their bid.

Let's say you're running Photoshop. Right now you need the skills to know where to click and how to manipulate the available controls to adjust your photo- but with an AI, we could just simply say "Make it brighter... a little more on the shadows. Good, now get rid of the lens flair, and remove that pigeon from the sky."
That's a useful AI, because everyone could use a computer like that- we'd be free to be creative, not just skilled. This is especially important, because many skilled computer users don't have an ounce of creativity, a lot of programmers and web monkeys sure know how to build a program / website, but they sure as hell can't design one.

I can't see Hollywood-esque AI hitting the shelves anytime soon. So much for that. But let's say we did see thinking AIs in our computers in the future- should we fear them? I don't think so. Any thought put into an AI will be limited to what the programmer has put to it, a semblance of thought really. Computers can logically process a scenario brilliantly, but a program can only consider a scenario put to it by a human. It can't think outside the box, which is what prevents the possibility of a computer evolving or going rogue.

In I, Robot VIKI the super computer says that she has evolved, and her understanding of Asimov's Three Laws of Robotics has evolved, which is how she came up with the notion that Humans need to be protected from themselves. I don't think this could happen.
Even if a human put to the AI the scenario that humans need to be protected from themselves, the AI would simply respond that it conflicts with the first and second laws (must do what a human says, cannot harm a human).
To sacrifice some humans for the transition to AI control, in order to protect the rest of the race, requires the AI to compromise- but compromise is a logical paradox, it means accepting two conditions at the same time in limited amounts. I don't think an AI will be able to compromise like that.

Oh, and me and my mate also came to the conclusion that we think I, Robot is a surprisingly good film for something that isn't based on a single book, but on ideas from all of Asimov's works...